RAG stands for retrieval-augmented generation. In ordinary language, it means the AI looks for relevant information in a set of documents or other sources before it writes an answer.
A general AI model may not know your current policies, contracts, manuals or private documents. RAG can help the system answer from material you provide.
The system can retrieve the wrong document, miss an important passage, use an old version, combine conflicting sources or write a conclusion that the retrieved text does not support.
Document source and version, chunk or passage lineage, retrieval quality, contradictory evidence, citations, access permissions and what the system does when it cannot find enough evidence.